Anthropic Warns Small AI Models Are Months From Self-Replication

Anthropic's red team says small AI models could self-replicate across insecure devices within months. Security researchers say the threat is real — but the defenses are already decades old.